Forecasting of Population

• The study of the characteristics of a population and of their evolution

through time and space constitutes the field of demography
• Population projection can be done by
– Trend models:
• treat the population as a whole i.e, without disaggregation with respect to
age, sex or other characteristics
• These methods take only time as independent variable
• Examples: linear model, exponential model, modified exponential model,
logistic model
– Composite Models:
• Treat the population as an aggregate of various groups. The evolution of
population results from the interactions from these groups.
• Example: cohort survival model
 Linear Trend Models
• The increase in equal time periods is constant.
i.e., the growth of population is linear with
time.
– Let the constant increase in population every year
is a and the population for year n is Pn
– then, P1 = Po+ a; P2 = P1+ a = Po+2a;
…. …
Pn = Po+na
– The practical method of calibration of a and Po is
by least square regression.
 Exponential Trend Model
• Exponential
– The increase in population in equal time periods is not
constant. The increase in population, (Pn+1 – Pn ), is
proportional to the present population, Pn
– i.e., Pn+1 – Pn = Pn; where,  is the growth rate.
P1 – P0 =  P0 or P1 = (1+ )P0
P2-P1 =  P1 or P2 = (1+ )P1 = (1+ )2P0
…………
Pn = (1+ )nP0
–  and P0 can be calibrated by least square regression
method after linearisng the above equation.
 Modified Exponential Model
• Characteristic of linear or exponential model is that the population level
continues to grow indefinitely. As this is unrealistic, models are proposed
based on the assumption of a finite limit to the population level.
• Premise: The remaining growth in population, i.e., the difference between
the final population level and the existing population level, is a constant
fraction of what it was at the previous time period.
i.e., (P∞-Pn)/(P∞-Pn-1) = 
 is a constant smaller than 1
P∞-P1 = (P∞-P0)
P∞-P2 = (P∞-P1) = 2(P∞-P0)
.
.
P∞-Pn = n(P∞-P0)
i.e., Pn = P∞- n(P∞-P0)
 Double Exponential Model
Assumption: The growth of the population is proportional to the population
level with a proportionality factor which instead of being constant increases
exponentially with time

(A)

at t = 0, from (A), a = P0 / P
 Logistic Trend Model
• The population initially grows moderately, picks up when the economic
base reaches a certain minimum level, and ultimately reaches a saturation
level. This cycle of population growth pattern is best depicted by a logistic
model.
• The rate of growth (t), as in exponential model, is not constant, but is a
linearly decreasing function of the demand level (Pt).
– i.e., t = a – b Pt (1)

but, (2)

therefore, (3)

The solution to the above differential equation is the logistic model

(4)

From (1), at t = ∞, t = 0 and therefore, P∞ = a/b

Calibration of Logistic Trend Model
• Generalized least squares or maximum likelihood method
can be used to calibrate the logistic model.
• Alternatively, the parameters a and b can be estimated by
plotting between (Pn+1 – Pn)/Pn and Pn. The intercept is a
and the absolute value of the slope is b.
(Pn+1-Pn)/Pn

b
a 1

Pn
 Comparative Models
• These models are also known as
– Ratio methods
– Market share models
• Typical Comparative Models